Towards Living Ontologies for Business Process Interoperability

Living Ontologies:
with applications to
Business Process Alignment and
Building Consensus
Peter Weinstein, PhD
Altarum Institute
March 28, 2006
Living Ontologies

A way to use ontologies designed to evolve

Ongoing opposing processes
– Differentiation: Users specialize terms for model accuracy
– Unification: Identify commonality with graph matching
Similarities
Core Concepts
Core Concepts
Generic Concepts
Organization-Specific
Concepts
Organization-Specific
Concepts
Original Models
Unified Model
Differences
Model unification creates a middle layer of shared concepts
www.altarum.org
AAAI SSS 06 - Semantic Web meets eGovernment
2
Unification Algorithm

A swarm intelligence approach

Concept agents seek matches that maximize similarity
– Based on lexical association and structural isomorphism
G: Create_RFQ
G: Vendor_Bid
A: Generate PTS
& eReq
A: Vendor Bid
C:Create PO
C:Prepare RFQ
D: Create RFQ
“Musical chairs”: when a concept moves it often kicks another out of its match
www.altarum.org
AAAI SSS 06 - Semantic Web meets eGovernment
3
Problem 1 –
Business Process Alignment

Want to analyze business processes for
interoperability or reengineering, but …

Semantic heterogeneity impedes comparison
Business process models can be hard to compare
www.altarum.org
AAAI SSS 06 - Semantic Web meets eGovernment
4
Solution Overview Business Process Alignment

Model processes on two levels
– Users work with familiar diagrams and other tools
– Internal representation with formal ontology

Unify the models
– An automatic process assisted by users (anytime, anywhere)

Compare processes
Process
Users interact with
problem-specific
models such as
process flow
diagrams
www.altarum.org
Swim
lane
AAAI SSS 06 - Semantic Web meets eGovernment
Flow
5
Comparison of Unified Models

Visualization of similarities and differences

Quantification of process alignment in [0, 1]
pink = similarities
blue/green = differences
A comparison visualization of manually unified models
www.altarum.org
AAAI SSS 06 - Semantic Web meets eGovernment
6
Initial Results

Experimental data
– Four purchasing processes for medium-sized manufacturers

Compared automatic to manual unification
– Current automatic results are “too good”
– Next step: richer multi-level data
Commonality Identified
0.7
0.6
0.5
Per Match
0.4
Avg. Portion
Properties 0.3
0.2
Shared
0.1
0
Manual baseline
0% confirmed
28000
24500
21000
17500
14000
10500
7000
3500
0
30% confirmed
Process Step
Automatic unification finds more commonality than exists in manually unified model
www.altarum.org
AAAI SSS 06 - Semantic Web meets eGovernment
7
Problem 2 –
Political Discourse

Consensus Builder will be a place on the
internet where people go to:
–Speak about things they know and care about
–Listen to others (if or when they are ready to listen)
–Be counted by a system that aggregates and
publishes beliefs
www.altarum.org
AAAI SSS 06 - Semantic Web meets eGovernment
8
Speaking
Consensus
Builder
SpeakingTo
to Consensus
Builder
Statement
Interpretation Quality
I am a 35 year old school administrator and
mother of two who has diabetes type 2.
Unfortunately, the insurance companies don’t
care about helping me protect my health. For
example, I am supposed to test my blood
sugar twice a day using test strips that cost
75 cents each. The insurance companies pay
for only one strip per day ...
Behaviors enabled without
further confirmation:
Very
Good
Quote in summaries
Vote on query-defined issues
Compare to other statements
Catalog and link statement
Index statement
Poor
User helps system
Submit
Simple Speak
interpret their statement
Clarifying
a Key
Term
Causality
Model
Dialogs
-- We have agreed that a key term in your
statement is health. Health can mean a
number of things. Let’s pick out the elements
that are important for your statement.
-- Help me with the insurance companies.
Can you be specific about what these are
and their connection to you?
-- What is the connection between test my
blood sugar and protect my health?
www.altarum.org
Need help!
Health - State of
hasHealth
Health History
hurts
American
Health Care
Quality
Behavioral health
Insurance
Family health
Mental health
neglects
Physical health
has Blood Pressure
Chronically
has Body Weight
ill
has (0..n) Disease(s)
AAAI SSS 06 - Semantic Web meets eGovernment
9
Listening
in Consensus
Listening
and Analysis Builder
Statement by Chaim54
There are many problems with health care in
America but let us not forget the important
contributions that health insurance
companies make to everyone’s welfare. Most
importantly, insurance spreads risk. Before
health insurance, falling ill with a disease
often meant financial ruin. Health insurance
companies also play an important role in
controlling costs
...
Compare
statements to
Simple Speak
mediate exchange
Analysis
Similarities
+ Insurance affects Health Care Quality
Differences (order by importance)
Replace private insurance with public
Improve insurance for preventive care
Good idea but doesn’t seem enough
+ American insurance hurts health care
+ American insurance helps health care
Don’t seem to be controlling costs well
Chronically ill are especially at risk
Causality Comparison
Comparison Aspects
-- Authoring Context (0.94 locality)
-- Causality (0.38 agreement)
Shared concepts on 1 of 3 levels of detail
Differences in relations
-- Terminology (0.82 concordance)
-- Timeline (0.35 agreement)
Shared actions on 1 of 3 levels of detail
Differences in timing
www.altarum.org
hurts
American Health
Insurance
Health Care
Quality
helps
neglects
Chronically
ill
AAAI SSS 06 - Semantic Web meets eGovernment
mitigates
Risk
controls
Health Care
Cost
10
LearningBe
from
Consensus Builder
Counted
Query
Stakeholder Status
Question: What kind of financial system
should America use to pay for health care ?
Stakeholders
Inner (magnify 1.5):
Chronically ill
Middle (magnify 3):
Health care providers
Outer:
Other Americans
A tool for learning
Simple Speak
Response
Green: Public Blue: Private Pink: Hybrid
Select display tool
- Chronically ill (30,389)
- Public insurance (10,326)
(947) Caring about health care. Frank231
(855) Try again. Yizkrit19
(123/42) Making Health Healthy. Leon36
(87/54) European health models. Molly 4
+ Private insurance (4,321)
+ Hybrid public/private solutions (5,742)
+ Health care providers (93,521)
+ Other Americans (912,827)
Orange: clarity Purple: wisdom
View
www.altarum.org
Organize
Specify display
Statement by Leon36
Title: Making Health Healthy
Nominations: 123 clarity; 42 wisdom
Endorsers (154); Disputants (72)
History: Exchanges with Mohammed1291,
Molly44, Chaim54 (endorser), Elizabeth932
The health insurance system in the United
States is undermined by two classic forms of
social dysfunction. These are called
Discounting the Future, and the Prisoner’s
Dilemma ...
Simple Speak
AAAI SSS 06 - Semantic Web meets eGovernment
11
Conclusions

Living Ontologies evolve through use
– Tolerate differences, maximize similarity
– Wrap agents around concepts to self-organize

Applications meet users where they work
– Ontologies belong under the hood

Benefits can include
– New scientific rigor for Business Process Reengineering
– Knowledge sharing to facilitate political discourse
www.altarum.org
AAAI SSS 06 - Semantic Web meets eGovernment
12